AI Rapidly Maps Landslides Post-Earthquake to Enhance Disaster Response Efforts

On April 3, 2024, a magnitude 7.4 earthquake—Taiwan’s strongest in 25 years—struck the country’s eastern coast. While stringent building codes prevented widespread structural damage, mountainous and remote regions suffered from severe landslides, exacerbating the disaster’s impact. In such situations, timely and accurate identification of affected areas is crucial for effective disaster response, and researchers are turning to artificial intelligence (AI) to enhance this process. Lorenzo Nava, a researcher jointly based at the Departments of Earth Sciences and Geography at the University of Cambridge, has developed an AI method to rapidly detect landslides from satellite imagery. After the Taiwan earthquake, Nava and his team used this AI to identify 7,000 landslides within three hours of acquiring the satellite images. This quick turnaround helps responders prioritize relief efforts in hard-to-reach areas. Landslides, often triggered by earthquakes or heavy rainfall, can be exacerbated by human activities like deforestation and construction on unstable slopes. They can lead to secondary hazards such as debris flows and flooding, making them particularly dangerous in certain environments. Nava’s research is part of a larger initiative at Cambridge called CoMHaz, led by Professor Maximillian Van Wyk de Vries, which focuses on understanding and predicting multihazards. The CoMHaz group integrates data from various sources, including satellite imagery, computer modeling, and fieldwork, to identify landslides and their causes. Nava’s AI method uses both optical satellite images and radar data, the latter of which can penetrate cloud cover and take images at night. Radar data, though challenging to interpret due to its greyscale depiction and distortion of landscape features, is ideal for AI-assisted analysis. In the Taiwan earthquake trial, the AI model successfully detected landslides that would have been hidden by clouds. However, Nava recognizes the need to enhance the model’s accuracy and transparency. He aims to build trust among decision-makers who may not be familiar with the technical aspects of AI algorithms. To achieve this, he is collaborating with the European Space Agency (ESA), the World Meteorological Organization (WMO), and the International Telecommunication Union’s AI for Good Foundation to develop an interpretable AI model. One of the group’s initiatives involves launching a data-science challenge to crowdsource improvements to the model. This challenge seeks contributions from the wider coding community to refine the AI’s functionality and incorporate features that explain its reasoning, such as visualizations showing the likelihood of landslides in specific areas. This approach ensures that the model’s outputs are not only accurate but also understandable and actionable for emergency response teams. Nava emphasizes the importance of transparency in high-stakes scenarios like disaster response: "Very often, the decision-makers are not the ones who developed the algorithm. AI can feel like a black box, and its internal logic is not always transparent, which can make people hesitant to act on its outputs. It’s crucial to make it easier for end users to evaluate the quality of AI-generated information before incorporating it into important decisions." To further illustrate the potential of this technology, Nava and Van Wyk de Vries have been working with local scientists and the Climate and Disaster Resilience in Nepal (CDRIN) consortium. They piloted an early warning system in Butwal, Nepal, a region vulnerable to landslides due to its unstable terrain. The system leverages AI to monitor and predict landslide risks, helping to protect communities and reduce the devastating effects of natural disasters. Industry insiders and experts commend Nava’s innovative approach, noting that AI-driven rapid landslide mapping can significantly enhance disaster response efforts. The Cambridge CoMHaz group and its international partners are paving the way for more resilient and prepared communities, especially in regions prone to natural hazards. As AI continues to evolve, its applications in disaster management hold the promise of saving lives and minimizing economic losses. The University of Cambridge and its collaborators are committed to advancing this technology, ensuring it not only meets scientific standards but also gains the trust and adoption of those on the front lines of disaster response. Through these efforts, they aim to create a more robust and reliable system that can be deployed anywhere in the world where landslides pose a threat.